dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

Unrolling the loops would be identical to what is already being done
in BLAS mode with -f blas as far as I can tell.

Or do you refer to the computation of the geometry tensor which is
performed also in BLAS mode?

/Anders


On Wed, Sep 13, 2006 at 10:19:17AM +0200, Garth N. Wells wrote:
> I've also had problems with very large .h files (some approaching 1GB which
> compilers cannot handle). What if FFC did not unroll the tensor contraction but
> used loops with a fixed size? Would this be easy from the FFC side? The compiler
> should be able to do some pretty good optimisations on this. Would it compare to
> FFC unrolling the loops?
> 
> Garth 
> 
> Quoting Anders Logg <logg@xxxxxxxxx>:
> 
> > I guess you mean the -O2 that you get with dolfin-config --cflags?
> > 
> > Perhaps we should avoid exporting optimization flags and debugging
> > flags to dolfin-config so these flags may be adjusted by users.
> > 
> > For now, just edit your dolfin-config script and remove the -O2. That
> > should take care of it.
> > 
> > Another option is to compile the forms with the option -f blas given
> > to FFC. This should generate much smaller h-files.
> > 
> > A warning: the -f blas option has not been tested for a while so I'm
> > not sure it works (it definitely does not work for forms that include
> > boundary integrals).
> > 
> > /Anders
> > 
> > 
> > On Tue, Sep 12, 2006 at 05:46:30PM +0200, Dag Lindbo wrote:
> > > Hi all,
> > > 
> > > Version 0.6.2-1 seems to be defaulting to some other optimization flags
> > > than 0.6.1 did. This is hurting compile-times to the extent that I can not
> > > use quadratic elements in my solvers (some h-files are rather big,
> > > ~2Mb)...
> > > 
> > > What's cusing the slowdown? How do I control what flags are used by
> > libtool?
> > > 
> > > Thanks,
> > > Dag Lindbo
> > > 
> > > _______________________________________________
> > > DOLFIN-dev mailing list
> > > DOLFIN-dev@xxxxxxxxxx
> > > http://www.fenics.org/mailman/listinfo/dolfin-dev
> > _______________________________________________
> > DOLFIN-dev mailing list
> > DOLFIN-dev@xxxxxxxxxx
> > http://www.fenics.org/mailman/listinfo/dolfin-dev
> > 
> _______________________________________________
> FFC-dev mailing list
> FFC-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/ffc-dev